High speed high accuracy motion detection and tracking on the parallel pipelined projection engine

Summary form only given. The use of the Radon transform in analyzing single image frames has been demonstrated to yield accurate and robust detection of straight edges in the presence of noise, variable light conditions, and various texture disturbances near those edges. The possibility of using similar ideas for high-accuracy detection and tracking of motion in the image is being explored. When tracking pure translation, the algorithm commences with an initialization phase. Projections are taken through a range of 180°, and a search in projection space for dominant linear features is performed. With a priori knowledge, this initial search can be reduced and a particular object of interest can be recognized in the projection domain. When tracking pure rotation, an initialization phase is again employed. For each subsequent image frame, M orientation angles are investigated, where M is the possible angular rotation between image frames divided by the desired resolution of angular orientation. M can be reduced by appropriate interpolation. The Radon transform is performed in real-time on an architecture called the parallel pipelined projection engine